Estimating Seasonal Behavior States from Biologging Sensor Data

Josh M. London
Devin S. Johnson, Paul B. Conn, Brett T. McClintock, Michael F. Cameron, and Peter L. Boveng

Alaska Fisheries Science Center, NOAA Fisheries Seattle, Washington, USA

14 December 2015

The Importance of Seasons

The seasonal timing of key, annual life history events is an important component of many species’ ecology.

  • Bird migrations coincide with availability of food en route
  • Salmon spawning timed with favorable water flow
  • Arctic ice seals linked with seasonal sea-ice

It’s All About the Timing

The timing of key life history events is well documented only for species found in accessible rookeries, breeding areas, or migratory corridors.

  • Pinniped rookery counts: first territory, first pup
  • Migration Timing: Eastern grey whales
  • Elephant seal molt season

Remote, Widely Disperssed Species a Challenge

Our knowledge of seasonal timing for species widely dispersed in inaccessible or remote habitats is poor. We only get a snapshot

  • Rare or irregular snapshots
  • e.g. a single aerial survey or rookery count

The Importance of Seasons

Seasonal periods important to marine mammals often do not align well with typical labels (i.e., spring, summer, winter, fall).

  • How can we identify species specific seasons?

Data from Biologgers

Could we use observations from bio-logging sensors to estimate seasons?

  • Fine-scale observations
  • Deployment lengths of several months, approaching 1 year
  • Records of both movement and behavior

Hidden Markov Models

  • Useful for pattern recognition in noisy time series data
  • Previously used with telemetry data to identify behavioral states
  • Behavioral states are, typically, shorter time frames
  • Duration in a state is not dependent on time already spent in a state

Multivariate Semi-Hidden Markov Models

Multivariate Semi-Hidden Markov Models

Hidden semi-Markov models allow an arbitrary sojourn distribution

R package mshmm

Regularly Spaced Observations

  • Argos/GPS-fastloc locations are irregular in time
  • Use movement model to predict regularly-spaced observations

R Package crawl — Continuous Time Correlated Random Walk

Predict locations 6-hourly to align with behavioral observations

Bearded, Ribbon and Spotted Seals

Predicted Movements

Haul-out Behavior - Bearded Seals

Haul-out Behavior - Ribbon Seals

Haul-out Behavior - Spotted Seals

Dive Behavior - Bearded Seals

Dive Behavior - Ribbon Seals

Dive Behavior - Spotted Seals

X-Y Displacement - Bearded Seals

X-Y Displacement - Bearded Seals

Seasonal States - Bearded Seals

Seasonal States - Bearded Seals

% Dry x Seasonal States

No. of Dives x Seasonal States

Bearing x Seasonal States

Spatial Distribution of Seasonal States

Seasonal States - Ribbon Seals

Seasonal States - Ribbon Seals

## List of 1
##  $ plot.title: list()
##   ..- attr(*, "class")= chr [1:2] "element_blank" "element"
##  - attr(*, "class")= chr [1:2] "theme" "gg"
##  - attr(*, "complete")= logi FALSE

Seasonal States - Ribbon

Seasonal States - Spotted Seals (Young of Year)

## List of 1
##  $ plot.title: list()
##   ..- attr(*, "class")= chr [1:2] "element_blank" "element"
##  - attr(*, "class")= chr [1:2] "theme" "gg"
##  - attr(*, "complete")= logi FALSE